Homology-based protein structure prediction: Fold recognition and alignment

We have developed a state-of-the-art program that recognizes structural similarities between an existing protein structure (template) and a given sequence (target), with no significant sequence identity (<20%) between the two. This method integrates a novel way of using multiple predicted secondary structures of the target, which minimizes the inaccuracies in any single prediction and finds the best match to a template. An automated and systematic way of enumerating variations within structural neighbors of a template allows us to find a better template built from fragments of existing protein structures. Results on a diversified target set, some of them being CASP targets, have shown significant improvement and the ability to build models with similar topology as the experimental structure of the target from a template that is only partially correct. The idea of using secondary structure and structural families is also applied to alignment between template and target with higher sequence identity, resulting in a smaller or similar RMSD than that from PDB-BLAST of the same template-target pair.